Quantitative structure-activity relationship model to predict the stability constant of Uranium coordination complexes for novel uranium adsorbent design

Quantitative structure-activity relationship model to predict the stability constant of Uranium coordination complexes for novel uranium adsorbent design | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Quantitative structure-activity relationship model to predict the stability constant of Uranium coordination complexes for novel uranium adsorbent design Hyun Kil Shin, Youngho Sihn This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4948478/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract A quantitative structure-activity relationship (QSAR) model for predicting the stability constant of uranium coordination complexes to accelerate the discovery of novel uranium adsorbents was developed and evaluated. Effective uranium adsorbents are crucial for mitigating environmental and health risks associated with uranium wastewater, an unavoidable byproduct of nuclear fuel production and power generation, as well as for sequestering uranium from seawater. QSAR modeling addresses the limitations of quantum mechanics calculations and offers a time- and cost-efficient computational approach for exploring vast chemical spaces. The QSAR model was built using a dataset of 108 uranium complexes, incorporating features such as physicochemical properties, coordination numbers of ligands, molecular charge, and the number of water molecules. Nineteen machine learning (ML) strategies were tested, and extreme gradient boosting (XGBoost) emerged as the best-performing ML algorithm, achieving an R² of 0.91 on the external test set after hyperparameter optimization. Including composition features significantly improved model performance, reflecting the physical factors influencing complex stability. Applicability domain analysis was conducted to evaluate model predictive performance. The QSAR model predicts stability constants from the molecular composition alone and is a valuable tool for the efficient design of safer and more sustainable uranium adsorption materials, potentially improving uranium collection processes. Physical sciences/Energy science and technology/Nuclear energy Physical sciences/Chemistry/Cheminformatics Uranium absorbent Quantitative structure-activity relationship Machine learning Stability constant Applicability domain Chemical space Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Uranium plays an indispensable role in meeting the world's growing energy demands, specifically in the production of nuclear energy. However, terrestrial uranium resources are finite, and as high-grade ores become increasingly scarce, the need for alternative sources of uranium is becoming more urgent. One such alternative is the extraction of uranium from seawater. The world's oceans contain approximately 4.5 billion tons of uranium, enough to meet the world’s uranium demand for over 10,000 years 1 . This immense resource, however, is distributed at a very low concentration of about 3.3 ppb 2 , making its extraction technically challenging. The retrieval of uranium is also associated with safety concerns since the traditional methods of uranium extraction, primarily terrestrial mining, pose significant environmental and health risks. Uranium mining and milling generate substantial radioactive waste, leading to contamination of water, soil, and air, with long-term consequences for ecosystems and human health. Thus, effective treatment methods are needed for the safe and sustainable extraction and use of uranium 3 . Uranium wastewater containing uranyl ions poses direct environmental and health hazards. With uranium wastewater being an unavoidable byproduct of nuclear fuel production and power generation, proper waste management is imperative. While adsorption is an effective method for sequestering uranium from wastewater 4 , the development of adsorbent materials that can efficiently capture dilute concentrations of uranium present in seawater, is a more difficult challenge. These materials must exhibit high selectivity for uranium over other metal ions, be resistant to biofouling, and maintain their performance over multiple adsorption-desorption cycles 5 . The extraction of uranium from seawater has been explored for decades, but most of the adsorbents were ineffective except polymeric adsorbents 6 and amidoxime-based materials 7 . Amidoxime-based polymers have emerged as the leading material for uranium adsorbents due to their strong affinity for the uranyl ion (UO₂²⁺). The amidoxime functional group, which forms stable complexes with uranium, is central to the ability of amidoxime-based adsorbents to selectively adsorb uranium from seawater. However, the large-scale implementation of seawater uranium extraction remains limited by the high costs associated with the low efficiency of uranium extraction. Therefore, new adsorbents which can efficiently extract uranium from seawater must be continuously explored and developed. The adsorption performance of uranium adsorbents can be assessed through the stability constant, which indicates the strength of the interaction between adsorbent material and uranium to form complexes 8 . The stability constant is represented as follows: $$\:\beta\:=\frac{\left[Complex\right]}{\left[Uranium\right]\left[Adsorbent\right]}$$ 1 where \(\:\beta\:\) is the stability constant, and [ Complex ] is the concentration of the uranium-adsorbent coordination complex, and [ Uranium ] and [ Adsorben t] are the concentrations of each species. Computational methods can be used to determine \(\:\beta\:\) . Quantum mechanics (QM) calculations can be used to predict \(\:\beta\:\) 9 , and QM methods have been used in material design often. However, QM methods cannot explore the vast chemical space quickly and cost-effectively. Consequently, the use of machine learning (ML) models is for novel material design is more efficient 10 . The quantitative structure activity relationship (QSAR) model is an ML model whose input is a representation of the molecular structure and whose output is the activity of the input molecule (i.e., experimentally measured properties). In the QSAR model, features calculated from molecular structures (also called descriptors), are used to predict the activity variation of the molecule as a consequence of structural variation. To the best of our knowledge, prior to this work, only one QSAR model for \(\:\beta\:\) prediction has been reported. Zahariev et al. developed a QSAR model based on graph neural network models and traditional ML models for predicting \(\:\beta\:\) for metal-ligand complexes with the aim of designing new selective ligands for the target metal ion 11 . The predictive accuracy of the QSAR model is heavily dependent on the applicability domain (AD) of the model 12 . The predicted result is reliable when the input molecule has similar structural features with training data. If the specific target molecules only take small portion in the entire training set, the structural pattern of such molecules are not well trained; therefore, there is a high possibility that the model can’t produce reliable prediction results on the molecule. Thus, for novel uranium adsorbent development, the developed model focuses on the chemical space of uranium complexes can provide better and reliable prediction results than the general model. In this study, we developed a QSAR model to predict \(\:\beta\:\) for uranium complexes. In total, 108 uranium complexes were collected with their stability constants. Descriptors used in the model are the physicochemical properties, coordination numbers according to ligand atom, charge number, and the number of water molecules due to hydroxylation. The molecular formula of the uranium complexes was used to calculate four physicochemical properties, specifically, water solubility, boiling point, melting point, and pyrolysis point using a neural network model for inorganic compounds. Among 19 ML models tested, extreme gradient boosting (XGBoost) achieved the best prediction performance in the external test set (R 2 : 0.91). Therefore, the evaluation results confirm that the model developed in this study is capable of discovering novel uranium adsorbents. Material and methods Data preparation The stability constant (log \(\:\beta\:\) ) was collected from research articles with the structure information of uranium coordination complexes (Supplementary information). The molecular formula and ligand atoms were used to represent the molecular structure of uranium coordination complexes. The total data size was 108, and the data set was divided into training and test sets with a ratio of 8:2; thus, the training set comprised 86 data points and the test set comprised 22 data points. Feature preparation Features were prepared based on the structural properties of the uranium coordination complexes such as the coordination number of each ligand (N, O, F, and Cl), charge, number of water molecules through hydroxylation, molecular weight, and four physicochemical properties (aqueous solubility, melting point, boiling point, and pyrolysis point). These four physicochemical properties were predicted based on the molecular formula of the uranium coordination complexes using a neural network model 13 . The physicochemical property prediction models used here were developed for inorganic compounds, in contrast to most existing models focusing on organic molecules. These models use the electron configuration of inorganic molecules based on the composition to calculate the four physicochemical properties. In the collected dataset, the coordination complexes have four different ligand atoms such as N, O, F, and Cl. Thus, the number of ligand atoms was used as a feature. Prepared features are available in supplementary tables (Training set: Table S1 , and external test set: Table S2 ) Machine learning model development Machine learning models were developed according to OECD QSAR validation guideline (Fig. 1 ) 12 . In order to reduce the grid search space for hyperparameter optimization, we used Automated ML (AutoML) to choose the best performing ML algorithm among 19 ML algorithms, namely, extreme gradient boosting (XGBoost), extra trees regressor (ET), gradient boosting regressor (GBR), Huber regressor, random forest regressor (RF), passive aggressive regressor (PAR), k-nearest regressor (k-NN), elastic net, Bayesian ridge, lasso least angle regression, lasso regression, AdaBoost regressor, ridge regression, linear regression, least angle regression, orthogonal matching pursuit, decision tree regressor, light gradient boosting machine (LightGBM), and dummy regressor. The best performing model was selected for further hyperparameter optimization. First, each ML algorithm was evaluated with 3-fold cross validation (CV) in PyCaret library version 3.3.2 14 . Second, the best performing algorithms from the first step were further optimized using Optuna library version 3.4.0 15 to find an optimum set of hyperparameters. ML models were developed with scikit-learn version 1.3.0 16 and xgboost library version 1.7.6 17 . The model performance was evaluated using the root mean squared error (RMSE) and r square (R 2 ). The model developed in this work was used to predict the \(\:\beta\:\) of candidate materials. Applicability domain analysis The QSAR model makes a valid prediction when the input molecule is sufficiently similar to the training set. Thus, it is crucial to perform AD analysis to determine if the prediction is valid. Basically, AD analysis is an outlier detection step. The simplest way to identify outliers involves checking the feature range. Any data whose feature value is out of the training set’s feature value range is considered an outlier. The outlier was identified using the leverage and warning values. The leverage value was calculated for the i- th compound ( \(\:{h}_{i}\) ) with the warning value for the model as below: $$\:{h}_{i}=\:{x}_{i}{\left({X}^{T}X\right)}^{-1}{x}_{i}^{T}$$ 2 $$\:{h}^{*}=\:\frac{3(p+1)}{n}$$ 3 where X is the feature matrix of the training set, \(\:{x}_{i}\) is a feature vector of i- th compound, \(\:p\) is the number of features used in the model, \(\:n\) is the number of compounds in the training set, and \(\:{h}^{*}\) is a warning value. If \(\:{h}_{i}\) is larger than \(\:{h}^{*}\) , then the compound is considered as an outlier. Then, the leverage value of the model was examined, and the outlier was identified using a William plot from the training and test sets. In candidate material prediction, the leverage value was compared with the warning leverage to check for outliers. Moreover, all the feature value ranges were examined, and the candidate material was considered an outlier if one of the features has a value outside the range of the values in training set. Results and discussion Dataset analysis The QSAR model prediction reliability was evaluated via AD analysis. This involves checking the chemical space which represents the structural diversity of the compounds in each dataset. Normally, the chemical space is visualized based on the molecular weight and the water/octanol partition coefficient (logP) for organic molecules; however, the logP model does not apply to inorganic molecules. In this study, four physicochemical properties were calculated from the molecular formulas of the uranium coordination complexes; therefore, the chemical space of the data set was compared based on the molecular weight and the four physicochemical properties. The chemical spaces of the training set, test set, and candidate materials were compared as shown in Fig. 2 . The training and test sets show a similar distribution in the chemical space. Also, the chemical space of the candidate material was similar to the training and test data sets; therefore, we can conclude that the model trained and validated with the dataset can be reliably used to predict the \(\:\:\beta\:\) of the candidate materials. Model development PyCaret was used to apply AutoML and test 19 ML models with 3-fold cross validation (Table 1). An ML model whose R 2 is higher than 0.5 from 3-fold CV produces highly reliable predictions. According to the internal validation result, the top 3 models were tree-based ensemble models. The fact that the linear models did not perform well implies that \(\:\beta\:\) is not linearly correlated with the features used in the model. Thus, advanced ML methods achieved good prediction accuracy. In this test, only XGBoost and ET showed good performance in internal validation (XGBoost R 2 : 0.539 / ET R 2 : 0.5). As XGBoost achieved the highest prediction accuracy, hyperparameter optimization was conducted for the XGBoost model and significant performance improvement was achieved (Fig. 3 ). The XGBoost model exhibited an R 2 of 0.91 and an RMSE of 6.032 for RMSE during external validation (Table 2). Feature importance was analyzed by developing models with and without composition features. When XGBoost was developed with five physicochemical properties (MW, logS, BP, MP, and PP) alone, the model only achieved 0.58 for R 2 on the external test set. When additional composition features were added such as the coordination number of ligands, charge of the molecules, and number of hydrogen molecules, the model performance was significantly improved (Table 3) The charge state of the uranium complex significantly impacts \(\:\beta\:\) . Higher charges typically result in stronger electrostatic interactions between the uranium ion and the ligands, leading to more stable complexes. This is consistent with the observed feature importance, where the inclusion of molecular charge as a feature notably improved the model performance. The coordination number, which reflects the number of ligand atoms bonded to the central uranium ion, directly influences the stability of the complex. Ligands with a high coordination number can donate more electrons to the uranium ion, stabilizing the complex through stronger bonding interactions. The sensitivity of the model to changes in coordination number underscores its critical role in determining complex stability. Additionally, the number of water molecules in the system is significant, as hydration can either stabilize or destabilize the complex depending on specific interactions with the central metal ion and surrounding ligands. These composition features, when incorporated into the model, allow for a more comprehensive representation of the physicochemical factors governing complex stability, thus explaining the significant improvement in model performance. Prediction result analysis The range of prediction values was examined, as the prediction values for the uranium complex should fall between the maximum and minimum values in the training set. According to the analysis comparing experimental and predicted results, the predicted values of \(\:\beta\:\) for the candidate materials were within the range of \(\:\beta\:\) in both the training and test sets (Fig. 4 ). AD analysis was further performed on the training and test sets (Fig. 5 ). The model domain contained 93% of the training set (6 molecules were considered as outliers), and only two molecules were out of the domain in the test set. Thus, the model developed in this study covered nearly all of the uranium coordination complexes in the training and test set. AD analysis was applied to the candidate materials. Even though the candidate materials were in the chemical spaces of the training and test sets, there is always uncertainty in the prediction values since the model was previously not exposed to the data. Therefore, we checked the feature ranges of the candidate material data and compared them to the training set. Even if the molecule was found in AD according to the leverage value, we checked the range of 11 features between the candidate data and the training set. If the molecule out of range even in a single feature, the molecule was also marked as an outlier. After leverage analysis, 9 molecules were found in the domain. Two of these had features out of the range of the training set values: the coordination number of N and aquatic solubility. As a result, 7 molecules were considered as reliably predicted by the model (Table 4 ). Detailed AD analysis can be found in Table S3. Table 4 AD analysis results for the candidate materials AD analysis in-domain out-of-domain Leverage 9 7 Feature range 11 5 Reliable prediction 7 Conclusions Nuclear energy will remain part of the world's sustainable energy mix, and uranium adsorption material development is paramount for protecting human and environmental health. Safer, effective, and cost-efficient novel materials for uranium adsorbents must be continuously developed. The QSAR model for stability constant ( \(\:\beta\:\) ) prediction was developed in this study to accelerate novel uranium adsorption material design. The QSAR model uses composition features and physicochemical properties to predict stable candidate materials. Given that the physicochemical properties can be calculated from the composition alone, appropriate candidate materials can be predicted prior to synthesis. This enhances the speed of candidate material discovery while reducing the cost. Among machine learning models, XGBoost achieved good performance in the test set. AD analysis can be applied to further eliminate outlier candidates and increase the predictive performance of the model. Declarations Competing interests The authors declare no competing interests. Author Contribution H.S. and Y.S. designed the concept of the paper together. Y.S. collected dataset and provided features for the model. H.S. analyzed dataset and trained the model. H.S. and Y.S. wrote and edited the manuscript together. H.S. prepared figures and tables. Acknowledgement This research was financially supported by the Ministry of Trade, Industry, and Energy (MOTIE) and Korea Institute for Advancement of Technology (KIAT) through the international cooperative R&D program (Project No. P0019147) Data Availability Data is provided within the supplementary information files. References Kim, J. et al. Recovery of Uranium from Seawater: A Review of Current Status and Future Research Needs. Sep. Sci. Technol. 48 , 367–387. 10.1080/01496395.2012.712599 (2013). Xie, Y. et al. Uranium extraction from seawater: material design, emerging technologies and marine engineering. Chem. Soc. Rev. 52 , 97–162. 10.1039/D2CS00595F (2023). Fan, M. et al. Review of biomass-based materials for uranium adsorption. J. Radioanal Nucl. Chem. 330 , 589–602. 10.1007/s10967-021-08003-4 (2021). Shabbir, S., Yang, N. & Wang, D. Enhanced uranium extraction from seawater: from the viewpoint of kinetics and thermodynamics. Nanoscale . 16 , 4937–4960. 10.1039/D3NR05905G (2024). Liu, P., An, M., He, T., Li, P. & Ma, F. Recent Advances in Antibiofouling Materials for Seawater-Uranium Extraction: A Review. Materials 16 (2023). Das, A. et al. Efficient Adsorption and Desorption of Uranium(VI) Using a Polymeric Adsorbent: A Combined Theoretical and Experimental Approach with Real-Life Alkaline Leach Liquor. Ind. Eng. Chem. Res. 63 , 5845–5862. 10.1021/acs.iecr.3c04314 (2024). Tang, N. et al. Amidoxime-based materials for uranium recovery and removal. J. Mater. Chem. A . 8 , 7588–7625. 10.1039/C9TA14082D (2020). Tamon, H., Mori, H., Ohyama, S. & Okazaki, M. Correlation of Adsorption Equilibrium of Uranium by Taking into Account Its Chemical Species in Seawater. J. Chem. Eng. Jpn . 23 , 433–438. 10.1252/jcej.23.433 (1990). Vukovic, S., Hay, B. P. & Bryantsev, V. S. Predicting Stability Constants for Uranyl Complexes Using Density Functional Theory. Inorg. Chem. 54 , 3995–4001. 10.1021/acs.inorgchem.5b00264 (2015). Axelrod, S. et al. Learning Matter: Materials Design with Machine Learning and Atomistic Simulations. Acc. Mater. Res. 3 , 343–357. 10.1021/accountsmr.1c00238 (2022). Zahariev, F. et al. Prediction of stability constants of metal–ligand complexes by machine learning for the design of ligands with optimal metal ion selectivity. J. Chem. l Phys. 160 , 042502. 10.1063/5.0176000 (2024). OECD. Guidance Document on the Validation of (Quantitative) Structure-Activity Relationship [(Q)SAR] Models. (2014). Shin, H. K. Electron configuration-based neural network model to predict physicochemical properties of inorganic compounds. RSC Adv. 10 , 33268–33278. 10.1039/D0RA05873D (2020). Ali, M. PyCaret: an open source, low-code machine learning library in python , (2020). https://www.pycaret.org Akiba, T., Sano, S., Yanase, T., Ohta, T. & Koyama, M. in Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining 2623–2631Association for Computing Machinery, Anchorage, AK, USA, (2019). Pedregosa, F. et al. Scikit-learn: machine learning in python. J. Mach. Learn. Res. 12 , 2825–2830 (2011). Chen, T. & Guestrin, C. in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining 785–794San Francisco, California, USA, (2016). Additional Declarations No competing interests reported. Supplementary Files supplementaryinfocuratedreferences.docx supplementarytables.xlsx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4948478","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":354281700,"identity":"c13ac8b2-238f-4b62-b80e-de743e586795","order_by":0,"name":"Hyun Kil Shin","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA/0lEQVRIie3RsWrCQBzH8V84cAqk4/8QzCtcKWgEHyYSsMtBOxUnCQi63AOc+BJOzieCLnmADB1aCs4ZHRx6JxW6XNPR4b4QAvfnQ/7hgFDoDhP2iUqRARSVaAzdBnkboSuJtCH6J4EjAIsN2smAZP9TvxLS9XzxNaqyWUJ5p2kwefGRoZaDx41dTLzvlk+yJuLKsJWGHJa+xWrZ5x+O0HjRlQ2ROJaMxZgK77/cSKotyRzZg7FLG3GLobYE9c9XAOkn1emNa0GxsISrivhK7eaREhM/ORZbri6jXqqfT3Q+zJKEij3O08JL8JBfX/GvI3tP8AMgMX8MQ6FQKOT6BlBuTxWqlJvOAAAAAElFTkSuQmCC","orcid":"","institution":"Korea Institute of Toxicology","correspondingAuthor":true,"prefix":"","firstName":"Hyun","middleName":"Kil","lastName":"Shin","suffix":""},{"id":354281701,"identity":"386402be-340c-4961-9935-c17f73fe556b","order_by":1,"name":"Youngho Sihn","email":"","orcid":"","institution":"Korea Atomic Energy Research Institute","correspondingAuthor":false,"prefix":"","firstName":"Youngho","middleName":"","lastName":"Sihn","suffix":""}],"badges":[],"createdAt":"2024-08-21 04:11:29","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4948478/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4948478/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":65460464,"identity":"5518cad6-ec7c-4c56-bf92-33a375899017","added_by":"auto","created_at":"2024-09-27 17:37:23","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1317191,"visible":true,"origin":"","legend":"\u003cp\u003eQSAR model development process. The composition of the molecule was used in feature preparation and physicochemical properties were calculated based on molecular formular of the compounds. The data set split ratio was 8:2 (training:test). The best machine learning algorithm was selected after 3-fold cross validation and was further developed through hyperparameter optimization.\u003c/p\u003e","description":"","filename":"figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-4948478/v1/63e768ffba0cca803874aa8e.png"},{"id":65460970,"identity":"e8d60555-bf01-440f-a785-fd5b2eff384d","added_by":"auto","created_at":"2024-09-27 17:45:23","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":488148,"visible":true,"origin":"","legend":"\u003cp\u003eChemical space of uranium coordination complexes visualized with molecular weight against aqueous solubility (A), melting point (B), boiling point (C), and pyrolysis point (D). The chemical space shows that training set and test set have similar structural patterns. Moreover, candidate materials also show similar chemical spaces, confirming that the model developed with training data and validated with test data can make reliable prediction on the candidate material.\u003c/p\u003e","description":"","filename":"figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-4948478/v1/1e378a64f81bf4ece255bcac.png"},{"id":65460460,"identity":"c2eea10b-2edb-4ed4-8fbd-fe82db40c432","added_by":"auto","created_at":"2024-09-27 17:37:23","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":89267,"visible":true,"origin":"","legend":"\u003cp\u003eParity plot of the best model. As it achieved high prediction accuracy, the parity plot was linear.\u003c/p\u003e","description":"","filename":"figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-4948478/v1/e7ea5d0a7f2b30852b7826ed.png"},{"id":65460465,"identity":"bcc01666-e12a-47d2-9436-3eb198123b90","added_by":"auto","created_at":"2024-09-27 17:37:23","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":400689,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of the experimental \u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;and the predicted \u0026nbsp;\u0026nbsp;. The experimental \u0026nbsp;\u0026nbsp;\u0026nbsp;from the training and test sets is visualized with the predicted \u0026nbsp;\u0026nbsp;\u0026nbsp;from the candidate materials (A). The training set (B) and test set (C) have a wider range of \u0026nbsp;\u0026nbsp;\u0026nbsp;values whereas the predicted \u0026nbsp;\u0026nbsp;\u0026nbsp;has a narrower range (D).\u003c/p\u003e","description":"","filename":"figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-4948478/v1/2a34dd1e6423c0bcdbcd2c6f.png"},{"id":65460461,"identity":"ec059e13-0adf-4596-98eb-cdfca7ecb686","added_by":"auto","created_at":"2024-09-27 17:37:23","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":137075,"visible":true,"origin":"","legend":"\u003cp\u003eWilliam plot for analyzing the applicability domain (AD) of the model. Only one compound was outside of the range of values of the test set, and 93% of compounds in the training set was in the AD domain (80 out of 86).\u003c/p\u003e","description":"","filename":"figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-4948478/v1/ae21544fc352bb36ab0ea07f.png"},{"id":67128788,"identity":"562d03c7-279c-4e98-9c9f-5f92df74a351","added_by":"auto","created_at":"2024-10-21 12:24:31","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2933566,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4948478/v1/8f5ab7e1-ebc0-41f9-8414-d5779d17cb4d.pdf"},{"id":65460969,"identity":"ac9951fa-8727-4985-a6e7-8430f1eca888","added_by":"auto","created_at":"2024-09-27 17:45:23","extension":"docx","order_by":7,"title":"","display":"","copyAsset":false,"role":"supplement","size":25981,"visible":true,"origin":"","legend":"","description":"","filename":"supplementaryinfocuratedreferences.docx","url":"https://assets-eu.researchsquare.com/files/rs-4948478/v1/2f5e9d3a357e9a66e148ddaa.docx"},{"id":65460466,"identity":"2f3fd7f0-5def-44fc-bb28-5c3313bc9e2b","added_by":"auto","created_at":"2024-09-27 17:37:23","extension":"xlsx","order_by":8,"title":"","display":"","copyAsset":false,"role":"supplement","size":26476,"visible":true,"origin":"","legend":"","description":"","filename":"supplementarytables.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-4948478/v1/bd5e24f722a3d9526cf9ec77.xlsx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Quantitative structure-activity relationship model to predict the stability constant of Uranium coordination complexes for novel uranium adsorbent design","fulltext":[{"header":"Introduction","content":"\u003cp\u003eUranium plays an indispensable role in meeting the world's growing energy demands, specifically in the production of nuclear energy. However, terrestrial uranium resources are finite, and as high-grade ores become increasingly scarce, the need for alternative sources of uranium is becoming more urgent. One such alternative is the extraction of uranium from seawater. The world's oceans contain approximately 4.5\u0026nbsp;billion tons of uranium, enough to meet the world\u0026rsquo;s uranium demand for over 10,000 years\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. This immense resource, however, is distributed at a very low concentration of about 3.3 ppb\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e, making its extraction technically challenging. The retrieval of uranium is also associated with safety concerns since the traditional methods of uranium extraction, primarily terrestrial mining, pose significant environmental and health risks. Uranium mining and milling generate substantial radioactive waste, leading to contamination of water, soil, and air, with long-term consequences for ecosystems and human health. Thus, effective treatment methods are needed for the safe and sustainable extraction and use of uranium\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. Uranium wastewater containing uranyl ions poses direct environmental and health hazards. With uranium wastewater being an unavoidable byproduct of nuclear fuel production and power generation, proper waste management is imperative.\u003c/p\u003e \u003cp\u003eWhile adsorption is an effective method for sequestering uranium from wastewater\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e, the development of adsorbent materials that can efficiently capture dilute concentrations of uranium present in seawater, is a more difficult challenge. These materials must exhibit high selectivity for uranium over other metal ions, be resistant to biofouling, and maintain their performance over multiple adsorption-desorption cycles\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. The extraction of uranium from seawater has been explored for decades, but most of the adsorbents were ineffective except polymeric adsorbents\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e and amidoxime-based materials\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. Amidoxime-based polymers have emerged as the leading material for uranium adsorbents due to their strong affinity for the uranyl ion (UO₂\u0026sup2;⁺). The amidoxime functional group, which forms stable complexes with uranium, is central to the ability of amidoxime-based adsorbents to selectively adsorb uranium from seawater. However, the large-scale implementation of seawater uranium extraction remains limited by the high costs associated with the low efficiency of uranium extraction. Therefore, new adsorbents which can efficiently extract uranium from seawater must be continuously explored and developed.\u003c/p\u003e \u003cp\u003eThe adsorption performance of uranium adsorbents can be assessed through the stability constant, which indicates the strength of the interaction between adsorbent material and uranium to form complexes\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. The stability constant is represented as follows:\u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\n$$\\:\\beta\\:=\\frac{\\left[Complex\\right]}{\\left[Uranium\\right]\\left[Adsorbent\\right]}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e is the stability constant, and [\u003cem\u003eComplex\u003c/em\u003e] is the concentration of the uranium-adsorbent coordination complex, and [\u003cem\u003eUranium\u003c/em\u003e] and [\u003cem\u003eAdsorben\u003c/em\u003et] are the concentrations of each species. Computational methods can be used to determine \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e. Quantum mechanics (QM) calculations can be used to predict \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e\u003csup\u003e9\u003c/sup\u003e, and QM methods have been used in material design often. However, QM methods cannot explore the vast chemical space quickly and cost-effectively. Consequently, the use of machine learning (ML) models is for novel material design is more efficient\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe quantitative structure activity relationship (QSAR) model is an ML model whose input is a representation of the molecular structure and whose output is the activity of the input molecule (i.e., experimentally measured properties). In the QSAR model, features calculated from molecular structures (also called descriptors), are used to predict the activity variation of the molecule as a consequence of structural variation. To the best of our knowledge, prior to this work, only one QSAR model for \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e prediction has been reported. Zahariev et al. developed a QSAR model based on graph neural network models and traditional ML models for predicting \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e for metal-ligand complexes with the aim of designing new selective ligands for the target metal ion\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. The predictive accuracy of the QSAR model is heavily dependent on the applicability domain (AD) of the model\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. The predicted result is reliable when the input molecule has similar structural features with training data. If the specific target molecules only take small portion in the entire training set, the structural pattern of such molecules are not well trained; therefore, there is a high possibility that the model can\u0026rsquo;t produce reliable prediction results on the molecule. Thus, for novel uranium adsorbent development, the developed model focuses on the chemical space of uranium complexes can provide better and reliable prediction results than the general model.\u003c/p\u003e \u003cp\u003eIn this study, we developed a QSAR model to predict \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e for uranium complexes. In total, 108 uranium complexes were collected with their stability constants. Descriptors used in the model are the physicochemical properties, coordination numbers according to ligand atom, charge number, and the number of water molecules due to hydroxylation. The molecular formula of the uranium complexes was used to calculate four physicochemical properties, specifically, water solubility, boiling point, melting point, and pyrolysis point using a neural network model for inorganic compounds. Among 19 ML models tested, extreme gradient boosting (XGBoost) achieved the best prediction performance in the external test set (R\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e: 0.91). Therefore, the evaluation results confirm that the model developed in this study is capable of discovering novel uranium adsorbents.\u003c/p\u003e"},{"header":"Material and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData preparation\u003c/h2\u003e \u003cp\u003eThe stability constant (log\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e) was collected from research articles with the structure information of uranium coordination complexes (Supplementary information). The molecular formula and ligand atoms were used to represent the molecular structure of uranium coordination complexes. The total data size was 108, and the data set was divided into training and test sets with a ratio of 8:2; thus, the training set comprised 86 data points and the test set comprised 22 data points.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eFeature preparation\u003c/h2\u003e \u003cp\u003eFeatures were prepared based on the structural properties of the uranium coordination complexes such as the coordination number of each ligand (N, O, F, and Cl), charge, number of water molecules through hydroxylation, molecular weight, and four physicochemical properties (aqueous solubility, melting point, boiling point, and pyrolysis point). These four physicochemical properties were predicted based on the molecular formula of the uranium coordination complexes using a neural network model\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. The physicochemical property prediction models used here were developed for inorganic compounds, in contrast to most existing models focusing on organic molecules. These models use the electron configuration of inorganic molecules based on the composition to calculate the four physicochemical properties. In the collected dataset, the coordination complexes have four different ligand atoms such as N, O, F, and Cl. Thus, the number of ligand atoms was used as a feature. Prepared features are available in supplementary tables (Training set: Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e, and external test set: Table \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e)\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eMachine learning model development\u003c/h2\u003e \u003cp\u003eMachine learning models were developed according to OECD QSAR validation guideline (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e)\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. In order to reduce the grid search space for hyperparameter optimization, we used Automated ML (AutoML) to choose the best performing ML algorithm among 19 ML algorithms, namely, extreme gradient boosting (XGBoost), extra trees regressor (ET), gradient boosting regressor (GBR), Huber regressor, random forest regressor (RF), passive aggressive regressor (PAR), k-nearest regressor (k-NN), elastic net, Bayesian ridge, lasso least angle regression, lasso regression, AdaBoost regressor, ridge regression, linear regression, least angle regression, orthogonal matching pursuit, decision tree regressor, light gradient boosting machine (LightGBM), and dummy regressor. The best performing model was selected for further hyperparameter optimization.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFirst, each ML algorithm was evaluated with 3-fold cross validation (CV) in PyCaret library version 3.3.2\u003csup\u003e14\u003c/sup\u003e. Second, the best performing algorithms from the first step were further optimized using Optuna library version 3.4.0\u003csup\u003e15\u003c/sup\u003e to find an optimum set of hyperparameters. ML models were developed with scikit-learn version 1.3.0\u003csup\u003e16\u003c/sup\u003e and xgboost library version 1.7.6\u003csup\u003e17\u003c/sup\u003e. The model performance was evaluated using the root mean squared error (RMSE) and r square (R\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e). The model developed in this work was used to predict the \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e of candidate materials.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eApplicability domain analysis\u003c/h2\u003e \u003cp\u003eThe QSAR model makes a valid prediction when the input molecule is sufficiently similar to the training set. Thus, it is crucial to perform AD analysis to determine if the prediction is valid. Basically, AD analysis is an outlier detection step. The simplest way to identify outliers involves checking the feature range. Any data whose feature value is out of the training set\u0026rsquo;s feature value range is considered an outlier. The outlier was identified using the leverage and warning values. The leverage value was calculated for the i-\u003cem\u003eth\u003c/em\u003e compound (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{h}_{i}\\)\u003c/span\u003e\u003c/span\u003e) with the warning value for the model as below:\u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e\n$$\\:{h}_{i}=\\:{x}_{i}{\\left({X}^{T}X\\right)}^{-1}{x}_{i}^{T}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equ3\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ3\" name=\"EquationSource\"\u003e\n$$\\:{h}^{*}=\\:\\frac{3(p+1)}{n}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e3\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cem\u003eX\u003c/em\u003e is the feature matrix of the training set, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{i}\\)\u003c/span\u003e\u003c/span\u003e is a feature vector of i-\u003cem\u003eth\u003c/em\u003e compound, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\)\u003c/span\u003e\u003c/span\u003e is the number of features used in the model, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:n\\)\u003c/span\u003e\u003c/span\u003e is the number of compounds in the training set, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{h}^{*}\\)\u003c/span\u003e\u003c/span\u003e is a warning value. If \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{h}_{i}\\)\u003c/span\u003e\u003c/span\u003e is larger than \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{h}^{*}\\)\u003c/span\u003e\u003c/span\u003e, then the compound is considered as an outlier. Then, the leverage value of the model was examined, and the outlier was identified using a William plot from the training and test sets. In candidate material prediction, the leverage value was compared with the warning leverage to check for outliers. Moreover, all the feature value ranges were examined, and the candidate material was considered an outlier if one of the features has a value outside the range of the values in training set.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results and discussion","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eDataset analysis\u003c/h2\u003e \u003cp\u003eThe QSAR model prediction reliability was evaluated via AD analysis. This involves checking the chemical space which represents the structural diversity of the compounds in each dataset. Normally, the chemical space is visualized based on the molecular weight and the water/octanol partition coefficient (logP) for organic molecules; however, the logP model does not apply to inorganic molecules. In this study, four physicochemical properties were calculated from the molecular formulas of the uranium coordination complexes; therefore, the chemical space of the data set was compared based on the molecular weight and the four physicochemical properties. The chemical spaces of the training set, test set, and candidate materials were compared as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The training and test sets show a similar distribution in the chemical space. Also, the chemical space of the candidate material was similar to the training and test data sets; therefore, we can conclude that the model trained and validated with the dataset can be reliably used to predict the\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e of the candidate materials.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003eModel development\u003c/h2\u003e \u003cp\u003ePyCaret was used to apply AutoML and test 19 ML models with 3-fold cross validation (Table\u0026nbsp;1). An ML model whose R\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e is higher than 0.5 from 3-fold CV produces highly reliable predictions. According to the internal validation result, the top 3 models were tree-based ensemble models. The fact that the linear models did not perform well implies that \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e is not linearly correlated with the features used in the model. Thus, advanced ML methods achieved good prediction accuracy. In this test, only XGBoost and ET showed good performance in internal validation (XGBoost R\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e: 0.539 / ET R\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e: 0.5). As XGBoost achieved the highest prediction accuracy, hyperparameter optimization was conducted for the XGBoost model and significant performance improvement was achieved (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The XGBoost model exhibited an R\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e of 0.91 and an RMSE of 6.032 for RMSE during external validation (Table\u0026nbsp;2).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFeature importance was analyzed by developing models with and without composition features. When XGBoost was developed with five physicochemical properties (MW, logS, BP, MP, and PP) alone, the model only achieved 0.58 for R\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e on the external test set. When additional composition features were added such as the coordination number of ligands, charge of the molecules, and number of hydrogen molecules, the model performance was significantly improved (Table\u0026nbsp;3)\u003c/p\u003e \u003cp\u003eThe charge state of the uranium complex significantly impacts \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e. Higher charges typically result in stronger electrostatic interactions between the uranium ion and the ligands, leading to more stable complexes. This is consistent with the observed feature importance, where the inclusion of molecular charge as a feature notably improved the model performance. The coordination number, which reflects the number of ligand atoms bonded to the central uranium ion, directly influences the stability of the complex. Ligands with a high coordination number can donate more electrons to the uranium ion, stabilizing the complex through stronger bonding interactions. The sensitivity of the model to changes in coordination number underscores its critical role in determining complex stability. Additionally, the number of water molecules in the system is significant, as hydration can either stabilize or destabilize the complex depending on specific interactions with the central metal ion and surrounding ligands. These composition features, when incorporated into the model, allow for a more comprehensive representation of the physicochemical factors governing complex stability, thus explaining the significant improvement in model performance.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003ePrediction result analysis\u003c/h2\u003e \u003cp\u003eThe range of prediction values was examined, as the prediction values for the uranium complex should fall between the maximum and minimum values in the training set. According to the analysis comparing experimental and predicted results, the predicted values of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e for the candidate materials were within the range of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e in both the training and test sets (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). AD analysis was further performed on the training and test sets (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). The model domain contained 93% of the training set (6 molecules were considered as outliers), and only two molecules were out of the domain in the test set. Thus, the model developed in this study covered nearly all of the uranium coordination complexes in the training and test set. AD analysis was applied to the candidate materials. Even though the candidate materials were in the chemical spaces of the training and test sets, there is always uncertainty in the prediction values since the model was previously not exposed to the data. Therefore, we checked the feature ranges of the candidate material data and compared them to the training set. Even if the molecule was found in AD according to the leverage value, we checked the range of 11 features between the candidate data and the training set. If the molecule out of range even in a single feature, the molecule was also marked as an outlier. After leverage analysis, 9 molecules were found in the domain. Two of these had features out of the range of the training set values: the coordination number of N and aquatic solubility. As a result, 7 molecules were considered as reliably predicted by the model (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Detailed AD analysis can be found in Table S3.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAD analysis results for the candidate materials\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAD analysis\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ein-domain\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eout-of-domain\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeverage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFeature range\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eReliable prediction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusions","content":"\u003cp\u003eNuclear energy will remain part of the world's sustainable energy mix, and uranium adsorption material development is paramount for protecting human and environmental health. Safer, effective, and cost-efficient novel materials for uranium adsorbents must be continuously developed. The QSAR model for stability constant (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e) prediction was developed in this study to accelerate novel uranium adsorption material design. The QSAR model uses composition features and physicochemical properties to predict stable candidate materials. Given that the physicochemical properties can be calculated from the composition alone, appropriate candidate materials can be predicted prior to synthesis. This enhances the speed of candidate material discovery while reducing the cost. Among machine learning models, XGBoost achieved good performance in the test set. AD analysis can be applied to further eliminate outlier candidates and increase the predictive performance of the model.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eCompeting interests\u003c/h2\u003e \u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eH.S. and Y.S. designed the concept of the paper together. Y.S. collected dataset and provided features for the model. H.S. analyzed dataset and trained the model. H.S. and Y.S. wrote and edited the manuscript together. H.S. prepared figures and tables.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThis research was financially supported by the Ministry of Trade, Industry, and Energy (MOTIE) and Korea Institute for Advancement of Technology (KIAT) through the international cooperative R\u0026amp;D program (Project No. P0019147)\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eData is provided within the supplementary information files.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eKim, J. et al. Recovery of Uranium from Seawater: A Review of Current Status and Future Research Needs. \u003cem\u003eSep. Sci. Technol.\u003c/em\u003e \u003cb\u003e48\u003c/b\u003e, 367\u0026ndash;387. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1080/01496395.2012.712599\u003c/span\u003e\u003cspan address=\"10.1080/01496395.2012.712599\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXie, Y. et al. Uranium extraction from seawater: material design, emerging technologies and marine engineering. \u003cem\u003eChem. Soc. Rev.\u003c/em\u003e \u003cb\u003e52\u003c/b\u003e, 97\u0026ndash;162. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1039/D2CS00595F\u003c/span\u003e\u003cspan address=\"10.1039/D2CS00595F\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFan, M. et al. Review of biomass-based materials for uranium adsorption. \u003cem\u003eJ. Radioanal Nucl. Chem.\u003c/em\u003e \u003cb\u003e330\u003c/b\u003e, 589\u0026ndash;602. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10967-021-08003-4\u003c/span\u003e\u003cspan address=\"10.1007/s10967-021-08003-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShabbir, S., Yang, N. \u0026amp; Wang, D. Enhanced uranium extraction from seawater: from the viewpoint of kinetics and thermodynamics. \u003cem\u003eNanoscale\u003c/em\u003e. \u003cb\u003e16\u003c/b\u003e, 4937\u0026ndash;4960. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1039/D3NR05905G\u003c/span\u003e\u003cspan address=\"10.1039/D3NR05905G\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, P., An, M., He, T., Li, P. \u0026amp; Ma, F. Recent Advances in Antibiofouling Materials for Seawater-Uranium Extraction: A Review. \u003cem\u003eMaterials\u003c/em\u003e \u003cb\u003e16\u003c/b\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDas, A. et al. Efficient Adsorption and Desorption of Uranium(VI) Using a Polymeric Adsorbent: A Combined Theoretical and Experimental Approach with Real-Life Alkaline Leach Liquor. \u003cem\u003eInd. Eng. Chem. Res.\u003c/em\u003e \u003cb\u003e63\u003c/b\u003e, 5845\u0026ndash;5862. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1021/acs.iecr.3c04314\u003c/span\u003e\u003cspan address=\"10.1021/acs.iecr.3c04314\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTang, N. et al. Amidoxime-based materials for uranium recovery and removal. \u003cem\u003eJ. Mater. Chem. A\u003c/em\u003e. \u003cb\u003e8\u003c/b\u003e, 7588\u0026ndash;7625. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1039/C9TA14082D\u003c/span\u003e\u003cspan address=\"10.1039/C9TA14082D\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTamon, H., Mori, H., Ohyama, S. \u0026amp; Okazaki, M. Correlation of Adsorption Equilibrium of Uranium by Taking into Account Its Chemical Species in Seawater. \u003cem\u003eJ. Chem. Eng. Jpn\u003c/em\u003e. \u003cb\u003e23\u003c/b\u003e, 433\u0026ndash;438. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1252/jcej.23.433\u003c/span\u003e\u003cspan address=\"10.1252/jcej.23.433\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (1990).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVukovic, S., Hay, B. P. \u0026amp; Bryantsev, V. S. Predicting Stability Constants for Uranyl Complexes Using Density Functional Theory. \u003cem\u003eInorg. Chem.\u003c/em\u003e \u003cb\u003e54\u003c/b\u003e, 3995\u0026ndash;4001. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1021/acs.inorgchem.5b00264\u003c/span\u003e\u003cspan address=\"10.1021/acs.inorgchem.5b00264\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAxelrod, S. et al. Learning Matter: Materials Design with Machine Learning and Atomistic Simulations. \u003cem\u003eAcc. Mater. Res.\u003c/em\u003e \u003cb\u003e3\u003c/b\u003e, 343\u0026ndash;357. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1021/accountsmr.1c00238\u003c/span\u003e\u003cspan address=\"10.1021/accountsmr.1c00238\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZahariev, F. et al. Prediction of stability constants of metal\u0026ndash;ligand complexes by machine learning for the design of ligands with optimal metal ion selectivity. \u003cem\u003eJ. Chem. l Phys.\u003c/em\u003e \u003cb\u003e160\u003c/b\u003e, 042502. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1063/5.0176000\u003c/span\u003e\u003cspan address=\"10.1063/5.0176000\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOECD. Guidance Document on the Validation of (Quantitative) Structure-Activity Relationship [(Q)SAR] Models. (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShin, H. K. Electron configuration-based neural network model to predict physicochemical properties of inorganic compounds. \u003cem\u003eRSC Adv.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 33268\u0026ndash;33278. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1039/D0RA05873D\u003c/span\u003e\u003cspan address=\"10.1039/D0RA05873D\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAli, M. \u003cem\u003ePyCaret: an open source, low-code machine learning library in python\u003c/em\u003e, (2020). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.pycaret.org\u003c/span\u003e\u003cspan address=\"https://www.pycaret.org\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAkiba, T., Sano, S., Yanase, T., Ohta, T. \u0026amp; Koyama, M. in \u003cem\u003eProceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery \u0026amp; Data Mining\u003c/em\u003e 2623\u0026ndash;2631Association for Computing Machinery, Anchorage, AK, USA, (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePedregosa, F. et al. Scikit-learn: machine learning in python. \u003cem\u003eJ. Mach. Learn. Res.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 2825\u0026ndash;2830 (2011).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, T. \u0026amp; Guestrin, C. in \u003cem\u003eProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining\u003c/em\u003e 785\u0026ndash;794San Francisco, California, USA, (2016).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Uranium absorbent, Quantitative structure-activity relationship, Machine learning, Stability constant, Applicability domain, Chemical space","lastPublishedDoi":"10.21203/rs.3.rs-4948478/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4948478/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eA quantitative structure-activity relationship (QSAR) model for predicting the stability constant of uranium coordination complexes to accelerate the discovery of novel uranium adsorbents was developed and evaluated. Effective uranium adsorbents are crucial for mitigating environmental and health risks associated with uranium wastewater, an unavoidable byproduct of nuclear fuel production and power generation, as well as for sequestering uranium from seawater. QSAR modeling addresses the limitations of quantum mechanics calculations and offers a time- and cost-efficient computational approach for exploring vast chemical spaces. The QSAR model was built using a dataset of 108 uranium complexes, incorporating features such as physicochemical properties, coordination numbers of ligands, molecular charge, and the number of water molecules. Nineteen machine learning (ML) strategies were tested, and extreme gradient boosting (XGBoost) emerged as the best-performing ML algorithm, achieving an R\u0026sup2; of 0.91 on the external test set after hyperparameter optimization. Including composition features significantly improved model performance, reflecting the physical factors influencing complex stability. Applicability domain analysis was conducted to evaluate model predictive performance. The QSAR model predicts stability constants from the molecular composition alone and is a valuable tool for the efficient design of safer and more sustainable uranium adsorption materials, potentially improving uranium collection processes.\u003c/p\u003e","manuscriptTitle":"Quantitative structure-activity relationship model to predict the stability constant of Uranium coordination complexes for novel uranium adsorbent design","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-09-27 17:37:18","doi":"10.21203/rs.3.rs-4948478/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"775f5b34-1ccb-4580-bea4-a31d3188bd2a","owner":[],"postedDate":"September 27th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":37650240,"name":"Physical sciences/Energy science and technology/Nuclear energy"},{"id":37650241,"name":"Physical sciences/Chemistry/Cheminformatics"}],"tags":[],"updatedAt":"2024-10-21T12:23:42+00:00","versionOfRecord":[],"versionCreatedAt":"2024-09-27 17:37:18","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4948478","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4948478","identity":"rs-4948478","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}